consume very slow when Inflights to many about server

[thedevop]

@mochi-co , currently we store all QoS > 0 messages. Even if receive quota is exceeded:

server/server.go

Lines 971 to 993 in 624dde0

	if out.FixedHeader.Qos > 0 {
	i, err := cl.NextPacketID() // [MQTT-4.3.2-1] [MQTT-4.3.3-1]
	if err != nil {
	s.hooks.OnPacketIDExhausted(cl, pk)
	s.Log.Warn("packet ids exhausted", "error", err, "client", cl.ID, "listener", cl.Net.Listener)
	return out, packets.ErrQuotaExceeded
	}
	out.PacketID = uint16(i) // [MQTT-2.2.1-4]
	sentQuota := atomic.LoadInt32(&cl.State.Inflight.sendQuota)
	if ok := cl.State.Inflight.Set(out); ok { // [MQTT-4.3.2-3] [MQTT-4.3.3-3]
	atomic.AddInt64(&s.Info.Inflight, 1)
	s.hooks.OnQosPublish(cl, out, out.Created, 0)
	cl.State.Inflight.DecreaseSendQuota()
	}
	if sentQuota == 0 && atomic.LoadInt32(&cl.State.Inflight.maximumSendQuota) > 0 {
	out.Expiry = -1
	cl.State.Inflight.Set(out)
	return out, nil
	}
	}

With Expiry = -1, it also appears the message will never be removed from Inflight. This will also make cl.NextPacketID very slow and eventually fail.

Perhaps we can make the following changes:

• If QoS > 0, immediately check if the client quota has reached
  • If reached, drop message for this client and call OnQosDropped hook
  • If not, store the message